app.py

import torch

import gradio as gr
import pytube as pt
from transformers import pipeline

MODEL_NAME = "openai/whisper-large-v2"

device = 0 if torch.cuda.is_available() else "cpu"

pipe = pipeline(
    task="automatic-speech-recognition",
    model=MODEL_NAME,
    chunk_length_s=30,
    device=device,
)


all_special_ids = pipe.tokenizer.all_special_ids
transcribe_token_id = all_special_ids[-5]
translate_token_id = all_special_ids[-6]


def transcribe(microphone, task):
    file = microphone

    pipe.model.config.forced_decoder_ids = [[2, transcribe_token_id if task=="transcribe" else translate_token_id]]

    text = pipe(file)["text"]

    return warn_output + text


def _return_yt_html_embed(yt_url):
    video_id = yt_url.split("?v=")[-1]
    HTML_str = (
        f'<center> <iframe width="500" height="320" src="https://www.youtube.com/embed/{video_id}"> </iframe>'
        " </center>"
    )
    return HTML_str


def yt_transcribe(yt_url, task):
    yt = pt.YouTube(yt_url)
    html_embed_str = _return_yt_html_embed(yt_url)
    stream = yt.streams.filter(only_audio=True)[0]
    stream.download(filename="audio.mp3")

    pipe.model.config.forced_decoder_ids = [[2, transcribe_token_id if task=="transcribe" else translate_token_id]]

    text = pipe("audio.mp3")["text"]

    return html_embed_str, text


demo = gr.Blocks()

mf_transcribe = gr.Interface(
    fn=transcribe,
    inputs=[
        gr.Audio(source="microphone", type="filepath", optional=True),
        gr.Radio(["transcribe", "translate"], label="Task", default="transcribe"),
    ],
    outputs="text",
    layout="horizontal",
    theme="huggingface",
    title="Whisper Large V2: Transcribe Audio",
    description=(
        "Transcribe long-form microphone or audio inputs with the click of a button! Demo uses the"
        f" checkpoint [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and 🤗 Transformers to transcribe audio files"
        " of arbitrary length."
    ),
    allow_flagging="never",
)




demo.launch(enable_queue=True)